new-project
Creates a new project folder with git, docs, plans, tests, CI and agent rules from the talk's vision; the user picks the stack.
Skill by stage2dev
Install
Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.
npx agentmarket install skill/stage2dev-new-project
Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.
Get the weekly diff
The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.
About
Use cases: devops, agent-building
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
Permissions
What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.
- Tools
- No tools found
- Secrets
- None asked for
- Install scripts
- None found
Network hosts it talks to 1
codeload.github.com· found in install command
Files it touches
~/.claude/skills/new-project
Tools it pre-approves 4
From allowed-tools in SKILL.md: the agent may use these without asking you each time.
Bash(sh"${CLAUDE_PLUGIN_ROOT}/scripts/py"scripts/guarda.pyinfo)
What's inside
- Skill name
new-project- Description
- Creates a new project folder with git, docs, plans, tests, CI and agent rules from the talk's vision; the user picks the stack. Use for "new project", "start a project", "scaffold", "projeto novo", "criar projeto", "começar um projeto".
- License
- MIT
- Allowed tools
- Bash(sh, "${CLAUDE_PLUGIN_ROOT}/scripts/py", scripts/guarda.py, info)
- Instructions
- 75 lines in SKILL.md · view on GitHub
Before installing
- Pre-approves some shell commands
Files (5)
- SKILL.md
- README.md · templates/
- ci.yml · templates/
- env.example · templates/
- gitignore.txt · templates/
SKILL.md (first lines)
# New project
Context: !`sh "${CLAUDE_PLUGIN_ROOT}/scripts/py" scripts/guarda.py info`
Folder name (optional): $ARGUMENTS
The JSON above is the context. Talk to the user in its `language`, honoring `reply_style`. Name every document with `names` (keys `vision`, `prd`, `architecture`, `design`, `decisions`, `roadmap`, `backlog`, `agents`), never a literal name. `plans_dir` is the plans folder (default `.planning`); `tools` says what is installed. If `info` failed, say so and use defaults.
Rule: you never choose the stack. Every stack, test and CI choice comes from the user's answer. No dependency is installed, no remote is created, nothing is pushed or committed (only `git init`).
## Step 0 - Preconditions
1. A vision must exist: `docs/<names.vision>` in the current repo, or one the user points to (made by `/tune-up:talk`). If none: say so, suggest `/tune-up:talk` first, and stop unless the user insists (then ask for the one-sentence idea and write a short vision from it).
2. If talk mode is on (the hook blocks code writes), ask the user to say an exit phrase first (en: "let's build", "start building"; pt-BR: "pode construir", "comece a aplicar") and wait. Create no file before that.
3. Read the vision and, if present, the PRD (MVP, constraints, chosen tech). Anything already decided there is not asked again.
## Step 1 - Ask what is missing
Use `AskUserQuestion`, up to 4 questions per call, recommended option first, always with an "other" way out. Skip what the documents or `$ARGUMENTS` already answer.
- Folder name (kebab-case; default from the vision's name) and where to create it (default: a new subfolder of the current directory; it must not exist yet or must be empty).
- Type: web app, site/landing, API, automation, game, mobile.
- Front-end stack (if the type has a screen): common options such as React + TypeScript, Vue, Svelte, plain HTML/CSS/JS, or a game engine, plus "other".
- Back-end stack (if the type has a server side): common options such as Node + TypeScript, Python + FastAPI, Go, or none (static / managed backend), plus "other".
- Tests: which test tool per side (default: the stack's usual one) and whether to add end-to-end tests now (default no).
- CI: GitHub Actions minimal workflow, other, or none.
Which sides exist:
| Type | frontend/ | backend/ |
|---|---|---|
| web app | yes | yes |
| site / landing | yes | only if the user asks |
| API | no | yes |
| automation | no | yes (scripts and jobs) |
| game | yes (client) | only if the user asks |
| mobile | yes (app) | ask |
Show a summary (name, type, sides, stack, tests, CI) and ask "create it?" before writing anything.
## Step 2 - Create the skeleton
Work in the new folder. Everything the stack decides is marked `[STACK]` in the templates; fill every mark from the answers. Keep it minimal: manifest, one entry file and one smoke test per side; no dependency installed, no lockfile.
1. `git init -b main` (no remote, no commit, no push). If `-b` is not supported, `git init` then `git symbolic-ref HEAD refs/heads/main`.
2. `docs/`: one file per document from `${CLAUDE_PLUGIN_ROOT}/templates/project/` (`vision`, `prd`, `architecture`, `decisions`; `design` only if the type has a screen, as a stub that says to run `/tune-up:design`). File names come from `names`. Copy the vision (and PRD) from the talk if they already exist elsewhere; otherwise fill the template from the talk. Fill the stack in the "Stack" section of the architecture document (name + version asked of the user, or "latest at install time"). Translate headings to `language`; delete empty sections.
3. `<plans_dir>/`: `names.roadmap` and `names.backlog` from the same templates, with the MVP phases from the PRD (or one empty phase).
4. `frontend/` and `backend/` (only the sides from the table): the manifest with `lint` and `test` scripts/commands of the stack, `src/` with one entry file, and a tests folder (`tests/` or the stack's convention) holding one smoke test that passes without any service.
5. `.tune-up/`: `project.md` and `config.json` from `${CLAUDE_PLUGIN_ROOT}/templates/profile/` (same rules as the `profile` skill: fill all sections of `project.md` in the project language, "not defined" for the unknown; in `config.json` keep the contract keys, drop `_comment*` keys, set `project`, `language`, `plans_dir`, safe defaults). Fill the "Definition of done" table with the real lint/test commands. Do not run `trust`.
6. `<names.agents>` at the root from `${CLAUDE_PLUGIN_ROOT}/templates/project/agents.md` (it already carries the 450-line rule and the smallest-change rule; keep both). Fill commands, style, tests, plans folder.
7. `README.md` from `${CLAUDE_PLUGIN_ROOT}/skills/new-project/templates/README.md`.
8. `.gitignore` from `templates/gitignore.txt` (same folder), keeping only the blocks of the chosen stack and removing the comment lines that contain `[STACK]`.
9. `.env.example` from `templates/env.example`: variable NAMES only, never a value.
10. CI (if chosen): `.github/workflows/ci.yml` from `templates/ci.yml`; one job per existing side, running install, lint and test with the stack's commands. For another CI system, write its minimal equivalent.
Do not create `LICENSE`, `CONTRIBUTING` or any file not listed.
## Step 3 - Check
- `grep -rn "\[STACK\]\|{{" .` finds nothing (git internals excluded). Fix every hit.
- `python3 -c "import json; json.load(open('.tune-up/config.json'))"` passes (`python` where `python3` is missing).
- `.env.example` has no value that looks like a secret.
- Each code file is at or under 450 lines.
- Do not run the install, lint or test commands: nothing is installed yet.
## Step 4 - Knowledge graph (only if installed)
If `tools` says graphify is present: tell the user to run `/graphify .` to build the first map (it is a skill, not a plain command, so you do not run it); later refreshes use `graphify update .`. If it is not present, say nothing about it.
## Step 5 - Finish
Show the created tree (`find . -maxdepth 2 -not -path './.git/*' | sort`). List the commands the user must run themselves to install dependencies (from the stack, for example the package manager's install in `frontend/` and the virtualenv + install in `backend/`), then the first test run. Say what is left open (`[ASSUMPTION]` tags, empty sections, "not defined" in the profile) and that nothing was committed.
Next steps: `/tune-up:design` (if there is a screen), then `/tune-up:plan` for the first phase.From stage2dev/tune-up, MIT.
More from stage2dev/tune-up
- 0.567score
setup
stage2dev
First-run setup of tune-up - language, owner, reply style, kinds of use, user.json, optional tools (gh, claude-mem, fast-jev, Caveman, Ponytail, Graphify, ECC).
SkillAFreshness: AClaude Code +9 - 0.567score
design
stage2dev
Builds the design system (DESIGN.md with tokens) and a static HTML prototype made from it.
SkillAFreshness: AClaude Code +9 - 0.561score
evaluate
stage2dev
Evaluates a third-party repo, skill or plugin before importing it.
SkillAFreshness: AClaude Code +9 - 0.561score
handoff
stage2dev
Writes a HANDOFF.md plus a pasteable prompt to continue in a new session.
SkillAFreshness: AClaude Code +9 - 0.561score
investigate
stage2dev
Investigates a bug, pasted error, "stopped answering", failed booking or customer conversation.
SkillAFreshness: AClaude Code +9 - 0.561score
learn
stage2dev
Records a new rule in memory, project or user rules (or the plugin source, if set).
SkillAFreshness: AClaude Code +9 - 0.561score
migration
stage2dev
Applies a database migration/SQL to production with a safe protocol.
SkillAFreshness: AClaude Code +9 - 0.561score
part
stage2dev
Runs ONE Part of an already planned phase item by item, with Opus reviewers and a gate, then STOPS.
SkillAFreshness: AClaude Code +9 - 0.561score
profile
stage2dev
Creates or updates the project profile (.tune-up/project.md and config.json) by detecting git, stack, tests, database and deploy.
SkillAFreshness: AClaude Code +9 - 0.561score
research
stage2dev
Research that changes NOTHING.
SkillAFreshness: AClaude Code +9 - 0.561score
ship
stage2dev
Commits and opens the PR with an honest body.
SkillAFreshness: AClaude Code +9 - 0.561score
skills
stage2dev
Finds the right skill for a task in the tune-up skills catalog and loads only that one.
SkillAFreshness: AClaude Code +9
Signals
- Stars
- 0
- Forks
- 0
- Open issues
- 0
- Last commit (this folder)
- Oct 20, 2026 (8 days ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- MIT
- Repository
- github.com/stage2dev/tune-up
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from stage2dev
- 0.567score
design
stage2dev
Builds the design system (DESIGN.md with tokens) and a static HTML prototype made from it.
SkillAFreshness: AClaude Code +9 - 0.567score
setup
stage2dev
First-run setup of tune-up - language, owner, reply style, kinds of use, user.json, optional tools (gh, claude-mem, fast-jev, Caveman, Ponytail, Graphify, ECC).
SkillAFreshness: AClaude Code +9 - 0.561score
investigate
stage2dev
Investigates a bug, pasted error, "stopped answering", failed booking or customer conversation.
SkillAFreshness: AClaude Code +9 - 0.561score
handoff
stage2dev
Writes a HANDOFF.md plus a pasteable prompt to continue in a new session.
SkillAFreshness: AClaude Code +9 - 0.561score
evaluate
stage2dev
Evaluates a third-party repo, skill or plugin before importing it.
SkillAFreshness: AClaude Code +9 - 0.561score
talk
stage2dev
Talk mode - interview, brainstorm and persona debate that turns an idea into project documents, no code.
SkillAFreshness: AClaude Code +9
Cite this page
new-project skill. Agent Mart. https://agentmart.dev/skills/stage2dev-new-project (last updated 2026-10-24)